This is paired data. Each person has two responses: their before and after
responses.
If you have one record per person and their seatbelt wearing is recorded as:
before no=0/yes=1
after no=0/yes=1
and intervention group is coded
intervention no=0/yes=1
then you could use:
mcc after before if intervention == 1
mcc after before if intervention == 0
This will give you two tables that will estimate the odds ratio - the odds
of wearing a seatbelt after, versus the odds before. So an odds ratio of 2
in the intervention group would mean that the odds of people in this group
wearing a seat-belt in the post-intervention period were twice the odds of
wearing a seat-belt in the pre-intervention period.
This won't give you a test between these two tables, however.
To do this, try the following, I think it will work.
Reshape the data do that each person now has two records: their 'before'
record and their 'after' record. You'll need a variable to identify which
pair of records belong to the same person - personid.
Code the variables like this:
intervention no=0/yes=1
period before=0/after=1
seatbelt no=0/yes=1
Now run:
xi:clogit seatbelt i.intervention*i.period,group(personid)
you'll get:
beta(period) and beta(intervention*period)
no beta for intervention because it doesn't change for a person.
The p-value associated with the interaction term tests the hypothesis that
there is no difference in the period effects between the intervention
groups - i.e. changes in seat-belt wearing were the same in both groups.
If it is significant, however, then
the odds ratio in the non-intervention group will be exp(beta(period))
and in the intervention group it will be
exp(beta(period)+beta(intervention*period)).
These should be the same as you get using mcc above.
Kieran
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]]On Behalf Of Ricardo
> Ovaldia
> Sent: Thursday, 29 January 2004 5:21 AM
> To: [email protected]
> Subject: st: Comparing change in rates - frustrating problem, please
> help
>
>
> Dear Stata users,
>
> This is a basic stat question. In an intervention
> trial to increase seat belt use 328 adults were
> randomized to either a no intervention group(control)
> or an educational intervention. After a few months
> they were contacted and asked questions about their
> seat belt use. Here is the resulting data:
>
> Pecent seat belt use:
> intervention before after
> Yes 29% 58%
> No 40% 52%
>
> The randomization was such that the baseline for the
> two groups was NOT the same. How can I correctly test
> if the intervention worked given that the baselines
> are different and there is a ceiling effect (i.e.
> percents can not exceed 100%).
>
> Thank you very much,
> Ricardo.
>
>
>
>
>
> =====
> Ricardo Ovaldia, MS
> Statistician
> Oklahoma City, OK
>
> __________________________________
> Do you Yahoo!?
> Yahoo! SiteBuilder - Free web site building tool. Try it!
> http://webhosting.yahoo.com/ps/sb/
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
